Decadal climate predictions for climate services

Freja Vamborg

Max Planck Institute for MeteorologyHamburg, Germany

A framework for decadal climate predictionMiKlip project

Making skillful (and useful) predictions

„Roadmap“ for making decadal climate predictions part of a climate services portfolio

Overview

A framework for decadal climate predictionMiKlip project

Making skillful (and useful) predictions

„Roadmap“ for making decadal climate predictions part of a climate services portfolio

Overview

Decadal climate predictions Climate projections

• An estimate of the actual future evolution of climate

• Predicts anthropogenic change and climate variability

• Initial state based on observations of slow climate components (ocean, others)

• Hindcasts (retrospective predictions) allow to assess skill

• One potential future evolution of climate

• Contingent on emissions scenarios

• Initial state not considered essential

What are decadal climate predictions and how do they differ from climate projections?

Thanks to J. Marotzke, MPI Meteorology

What are decadal climate predictions and how do they differ from climate projections?

Daily Monthly Seasonal Decadal Multi-decadal to centennial

Initial value problem:

Forced boundary condition

Picture: E . Winqvist

Decadal climate predictions - a candidate for climate services?

YES•time-horizon appropriate for decision making •indication of predictability for certain quantities

– Atlantic hurricanes (Smith et al. 2010)– Arctic sea ice, heatwaves, impacts?

Decadal climate predictions - a candidate for climate services?

YES•time-horizon appropriate for decision making •indication of predictability for certain quantities

RECOMMENDATION •develop a system for decadal climate predictions (World Summit 2009)•“Decadal exchange“ (lead Hadley Center1)

1. Smith et al. (2012) submitted to Climate Dynamics

Decadal climate predictions - a candidate for climate services?

YES•time-horizon appropriate for decision making •indication of predictability for certain quantities

RECOMMENDATION •develop a system for decadal climate predictions (e.g. World Summit 2009)•“Decadal exchange“ (lead Hadley Center)

Key Question can predictability be established?MiKlip project1

1. www.fona-miklip.de

Research challenges

Only difference between hindcasts is the initialisation datasetRight panel - clear improvement over the tropics

Initialisation

BAD GOOD

Correlation with HadISST for lead year 2

Graphic from H. Pohlmann, MPI Meteorology

Only difference between hindcasts is the initialisation datasetRight panel - clear improvement over the tropics

Need: Initialisation techniques and state estimates

Initialisation

BAD GOOD

Correlation with HadISST for lead year 2

Graphic from H. Pohlmann, MPI Meteorology

Lack of robust decadal climate variabilty in global models•differences in the representation of processes e.g., AMOC•missing processes

Processes

Lack of robust decadal climate variabilty in global models

Need: increased process understanding

Processes

Increased relevance for usersBut, so far: •little focus on regional modelling + decadal climate prediction

Regionalisation

Graphics from Andreas Will (BTU Cottbus) and DKRZ

Little regional modelling + decadal climate prediction

Need to: – investigate key regions – combine global and regional models

Regionalisation

Graphics from Andreas Will (BTU Cottbus) and DKRZ

How do we judge what information is of use? •Scientific perspective:

– Skill (statistical)

•Societal perspective– Usefulness /applicability for decision making

Evaluation

How do we judge what information is of use? •Scientific perspective:

– Skill (statistical)

•Societal perspective– Usefulness /applicability for decision making

NEED to: •develop tools and methods to evaluate skill of decadal predictions•involve potential users

Evaluation

Ultimate challenge: system for decadal prediction

MiKlip project aim

Exchange of results during project development stages

Project coordination: J. Marotzke, D. Stammer, U. Cubasch, Ch. Kottmeier, U. Ulbrich, P. Becker

A framework for decadal climate predictionMiKlip project

Making skillful (and useful) predictions

„Roadmap“ for making decadal climate predictions part of a climate services portfolio

Overview

Current decadal prediction system“Baseline 0“

Global model: MPI-ESM-LR•Historical run: CMIP5 (all forcings) •Hindcasts: - 1961 – 2011

- initialised yearly- 3 ensemble members (10 for 5th year)

Global model: MPI-ESM-LR•Historical run: CMIP5 (all forcings) •Hindcasts: - 1961 – 2011

- initialised yearly- 3 ensemble members (10 for 5th year)

Skill measure for hindcasts•RMSE compared to

– historical runs– persistence

Current decadal prediction system“Baseline 0“

Initialised experiments compared to historical: skill in North Atlantic, reduced

skill in tropics, low skill over land

Müller et al. (2012) submitted to GRL

GOOD

BAD

Global mean surface T – lead year 2-5 RMSE w.r.t. HadISST & GHCN-CAMS

Prediction of quantities with known skill+ confidence in physical mechanism

Graphic from H. Pohlmann, MPI Meteorology

Prediction of quantities with known skill+ confidence in physical mechanism

• How relevant is this information from a climate services point of view?

Graphic from H. Pohlmann, MPI Meteorology

Boosting skill by reduced time-interval? – seasonal patterns in EuropeRMSE for Sept-Oct-Nov

Müller et al. (2012) submitted to GRL

Boosting skill by reduced time-interval? – seasonal patterns in EuropeRMSE for Sept-Oct-Nov

Müller et al. (2012) submitted to GRL

Boosting skill by reduced time-interval? – seasonal patterns in Europe

RMSE Sept-Oct-Nov in Eastern Europe

Müller et al. (2012) submitted to GRL

HindcastsPersistenceUninitialised

RSME lower for initialised runs, but not as consistent as for annual means

Boosting skill by reduced time-interval? – seasonal patterns in Europe

RMSE Sept-Oct-Nov in Eastern Europe

Müller et al. (2012) submitted to GRL

HindcastsPersistenceUninitialised

• Inter-decadal trend more realistic in regional modelpotential for increasing skill by using regional model contingent on global model skill in SSTs

Boosting skill by regionalisation ? 1961-99 summer rainfall trend over N Africa

Global (ECHAM5) Regional (REMO) Obs. (GPCC)

Graphic: Prof. H. Paeth and A. Paxian, Uni Würzburg

-1500 -200 - 20 20 200 1500 (mm/yr)

Baseline 0 Baseline 1 (new initialisation)

Baseline 0: decadal trend in Atlantic SST not capturedBaseline 1:

Positive correlation for the whole Atlantic basin SSTs necessary for n. Africa might be better capturedHighlights the need for several iterations within the project

Updating the decadal prediction system

With the global model: •robust skill for some regions (e.g. North Atlantic)•indication of increased skill over land when considering seasons With the regional models: •potential for added value, if skill in the global model can be established

Skill summary

With the global model: •robust skill for some regions (e.g. North Atlantic)•indication of increased skill over land when considering seasons With the regional models: •potential for added value, if skill in the global model can be established

Promising results, but a lot of work before being able to offer a „range“ of regions/times/quantities with skill

Need iterations involving all MiKlip modules

Skill summary

A framework for decadal climate predictionMiKlip project

Making skillful (and useful) predictions

„Roadmap“ for making decadal climate predictions part of a climate services portfolio

Overview

Roadmap1. Establish confidence in skill before involving users

Framework for decadal predictionImprove system skill (and understanding) from a scientific perspective This is where MiKlip (and others) are now

Decadal climate predictions as part of a climate services portfolio

Roadmap1. Establish confidence in skill before involving users Once potential for skill has been demonstrated: 2. Involve users in particular regions/with particular quantities of interest

– e.g., useful indices, seasons, quantities

Decadal climate predictions as part of a climate services portfolio

Roadmap1. Establish confidence in skill before involving users 2. Involve users in particular regions/with particular quantities of interest When users have been involved 3. Pin-point processes, regions and evaluation tools necessary for users‘ needs

improvement of skill system from a user/climate services perspective

Decadal climate predictions as part of a climate services portfolio

Roadmap1. Establish confidence in skill before involving users 2. Involve users in particular regions/with particular quantities of interest 3. Pin-point processes, regions and evaluation tools necessary for users‘ needs 4. Provide decadal predictions for climate services

Decadal climate predictions as part of a climate services portfolio

Roadmap1. Establish confidence in skill before involving users 2. Involve users in particular regions/with particular quantities of interest 3. Pin-point processes, regions and evaluation tools necessary for users‘ needs 4. Provide decadal predictions for climate services

Decadal climate predictions as part of a climate services portfolio